Home Technical Talk

Why so blue?

1
Alright, so most people have seen or created specular maps that have prominent blue colors in them, for various effects, but it seems many do no actually understand why we do this.

I've had a few discussions with programmers about this from time to time, because they never seem to understand why we would want anything but grayscale, so i figured i would give a few examples here and try to open up some discussion about colored spec in general. Generally these discussions end in the programmer telling me something about gamma which i dont really understand.

Now the most common use most people will say, is use blue for specular in your character textures. What they really mean is use the inverse color of your diffuse for your specular. Why do we do this? Its simple, adding light on top of a tan/brownish color is going to give you nasty yellow highlights.

You can see clearly in this image what i mean:

blueee.jpg

Specular in most 3d engines is done additively, so we can sort of test this out in photoshop.

The top two images are blue(left) and gray(right) using the normal blend mode in PS.

The bottom two images are the same, but using additive. You see how the gray image actually turns yellow, while the blue is a much cleaner near-white result. Its worth noting that if your additive color(spec color)
is bright enough, your end result is always going to be white, but the gradient to that point may still be yellow, and if the case of skin, you want very minimal specular reflection generally, so you wont have a very high spec value.

fingerslol.jpg
You can see in this ref here(my fingers lololol) that the highlight is more of a clean white than anything else, if you colorpick it, its a slight pink, but a quick look gives you the distinct impression of "white" more so than a yellow or red or anything else. Even in areas where the highlight is quite faint, its still distinctly white.

So when the goal is to get nice clean white highlights, it helps to use the inverse color of your diffuse.

Another good example that pretty much follows the same exact theory and usage is lacquered wood, the shiny wood you would see on the stock of a rifle for instance, tends to be a clean white or even bluish highlight in some cases. One of my biggest pet peeves is seeing people use grey, or even worse a yellow/brown spec color for wood, it just looks *terrible*.

here are some examples of what i mean:

1dypnt.jpg
http://www.fortunecity.com/olympia/kickbox/377/b19/b19stock1.jpg


Some other good uses for color in your specular are:

Bronze/Gold

Its good to have more of an orange/brown color in the diffuse, and a yellow/greenish color in the spec, to get this sort of effect:
2489931647_88fb822b32.jpg

Certain materials like christmas ornaments, or certain cloth(like satin) or types of plastic you may want to push the color of the diffuse into the spec to get an even more saturated highlight:
christmas_ornaments.jpg

Replies

  • Mechadus
    Wow, thanks EQ - very handy reference thread. Looks like you need to use a higher rez image for your kitchen floor tho - shits blurry! :P

    I learned something very much like this in HS physics which I use sometimes in spec maps... Basically invert the diffuse color to simulate how colored light is reflected off an object, thus canceling out 'wrong' colors.. Is this a good rule of thumb for in engine textures, or is this mostly a 3d animation trick? Just wondering because I also only really ever have seen blue on spec maps in my limited experience.

    -N!
  • EarthQuake
    Well the general rule here, is that if you want a less saturated highlight, use the inverse color, and if you what a more saturated highlight, use the same color. How you actually apply that is very specific to what material you're trying to create.
  • monkeyscience
    Offline / Send Message
    monkeyscience polycounter lvl 12
    There may be something to your first diagram EQ. IRL specular hilights aren't exactly additive because if incident light hits the oily surface on the skin and bounces right off, it will not illuminate the yellowish, diffuse flab underneath. Really the equation becomes SPEC + (1-SPEC)*DIFFUSE. Adding some inverse diffuse color to the spec map probably approximates this.

    Also the light source in the photo is not white, its blue, check out the "white" kitchen floor.
  • EarthQuake
    The shot was taken with natural, overcast lighting, which is about as neutral as i think its going to get. And how dare you call my floor white, you racist. While i'm sure there is a little blue being added from the sky, i dont think its a very large amount.
  • jocose
    Offline / Send Message
    jocose polycounter lvl 11
    This is really informative. Thanks a lot EQ.
  • SideEffect
    Offline / Send Message
    SideEffect polycounter lvl 19
  • EarthQuake
    ok white balanced on my camera to be as accurate to what the floor actually looks like(slight yellow/offwhite color) and a darker exposure to see the detail a bit more clearly. I think this still paints the same picture, spec = clean white. =)
    fingerslol2.jpg
  • oXYnary
    Offline / Send Message
    oXYnary polycounter lvl 18
    EQ, if your engine only allows greyscale specs in a alpha. But you can set the color of the spec via umm whats it called. Not CMYK, the 4 numeric system manually in the shader setup. Will this still help? As per since the spec color wont change, and if the texture is across multiple surface types.

    I mean is there a neutral value? Or should I make a copy of the final texture with a average blur filter to see what the most prevalent color is and make the 4 digit the opposite of that?
  • warby
    Offline / Send Message
    warby polycounter lvl 18
    this is great information i was often wondering why the blue .... however:

    lets not forget than if you use a grey scale specular mask you can get away with just "1" channel !!! j
    ust stash the spec mask into the alpha channel of your normal map or defuse map.
    if 99% of the shaders in your game use specular masks than that is a HUGE amount of texture memory !
    that slight color difference, that is going to get raped by all the bloom and color correction post filters anyway, might not matter that much compared to texture variety and resolution!
  • MoP
    Offline / Send Message
    MoP polycounter lvl 18
    warby, actually in general if you put a specular mask into the alpha then you're not saving memory at all, in most common cases.
    Since DXT1 (3 channels) is half the size of DXT5 (4 channels), so you might as well have two DXT1 textures instead of a single DXT5, it adds up to the same amount of texture memory at runtime and you get colour for "free" in your specular.
  • EarthQuake
    MoP wrote: »
    warby, actually in general if you put a specular mask into the alpha then you're not saving memory at all, in most common cases.
    Since DXT1 (3 channels) is half the size of DXT5 (4 channels), so you might as well have two DXT1 textures instead of a single DXT5, it adds up to the same amount of texture memory at runtime and you get colour for "free" in your specular.

    Absolutely.

    Also i've been told that a single8 bit image cannot be compressed by the graphics card, so while your grayscale spec map is 1/2 the size of your rgb map, compressed that rgb map is 1/4th the size. Thus actually being MORE expensive to use grayscale spec on its own.
  • EarthQuake
    oXYnary wrote: »
    EQ, if your engine only allows greyscale specs in a alpha. But you can set the color of the spec via umm whats it called. Not CMYK, the 4 numeric system manually in the shader setup. Will this still help? As per since the spec color wont change, and if the texture is across multiple surface types.

    I mean is there a neutral value? Or should I make a copy of the final texture with a average blur filter to see what the most prevalent color is and make the 4 digit the opposite of that?

    I'm not exactly sure what your asking here, seems you may be thinking a bit too hard about it? =D

    If you HAVE to use grayscale, its good if your shader supports and RGB value to tint the entire image, you'll just have to tweak this until you get a good result, because you may be balancing flesh and non flesh in the same image.
  • jeffdr
    Offline / Send Message
    jeffdr polycounter lvl 11
    Really the equation becomes SPEC + (1-SPEC)*DIFFUSE
    This is not correct, though it sort of is in spirit. The actual relationship is more complicated, as it depends on the incident angle of the light source. You can have for example angles where the viewer sees no specular, but the surface is reflecting a lot of light specularly, the result being dark diffuse but no visible spec. tl;dr - this equation is wrong, don't use it.

    Also fwiw I don't buy that the artists are just trying to emulate blue light by putting blue in the spec maps, I think they are actually on to something here, a flaw in gamma correction or the RGB color model or some other technical wankery.

    Also mechadus: you learned this in physics class?? If you know a physical basis for the phenomenon, then do share it please!
  • Ruz
    Offline / Send Message
    Ruz polycount lvl 666
    so I suppose that you would vary the spec colour a little ie slightly warmer colour for the lips.
    I personally found that it hard to avoid it looking like you painted the model with inks if you have too much color in the spec.
    it can look quite saturated I mean.

    If you haven't got a good grasp of warm and cool colors it can be a bit tricky to get your head around

    whilst testing I mainly use a grayscale spec.

    The idea for a blue spec may have come from the color of jacks lips in titanic:)
  • Ben Apuna
    Great thread EarthQuake! Finally someone put all this specular color info into a concise post. I knew about the invert color for skin from a previous polycount thread but now it all makes sense, thank you very much :)
  • warby
    Offline / Send Message
    warby polycounter lvl 18
    MoP wrote: »
    warby, actually in general if you put a specular mask into the alpha then you're not saving memory at all, in most common cases.
    Since DXT1 (3 channels) is half the size of DXT5 (4 channels), so you might as well have two DXT1 textures instead of a single DXT5, it adds up to the same amount of texture memory at runtime and you get colour for "free" in your specular.



    hmmmm now that IS interesting ...

    the truth is i take the memory efficiency thing so far i usually just use a channel from the dxt1 - diffuse texture do some contrast adjustments in the material/shader and hook that into the specular mask.

    because thats what most environment artists do anyway ... just taking the diffuse photo of a wall in photoshop and turning the contrast way up to make the spec mask and than call it a day ...
  • Bal
    Offline / Send Message
    Bal polycounter lvl 17
    Very interesting thread guys, and yeah I often heard the "use blue cause of the sky" explanation, which always sounded kind of silly.

    Sorry to go off topic a bit, but to come back to what warby and MoP are saying :
    Why is DXT5 2x times larger than DXT1 if it only has an additional channel?
    And I guess another advantage to just using grayscale specular, is even if you're using 2 DXT1 textures, you can use the 2 other channels of the specular texture for stuff like gloss and reflection maps (or even apha, to avoid using a DXT5 on the diffuse texture).
  • CrazyButcher
    Offline / Send Message
    CrazyButcher polycounter lvl 18
    why is dxt5 2x times larger?

    http://en.wikipedia.org/wiki/S3_Texture_Compression
    btw for normalmaps dxt1 is rather sucky quality, best quality is a swizzled dxt5. (and ofc 3dc)
  • warby
    Offline / Send Message
    warby polycounter lvl 18
    so you get something for your money with dxt 5 right ???
    dxt5s usually look less altered than dxt1 textures !? ( haven't used one in ages)
  • System
    Offline / Send Message
    System admin
    warby wrote: »
    so you get something for your money with dxt 5 right ???
    dxt5s usually look less altered than dxt1 textures !? ( haven't used one in ages)

    Sorry to jump in on your discussion but I used quite a few of these in conjunction with nvidia's direct x toolkit, simPE and Sims 2 engine. Yes DXT1 was poorish quality, DXT3 was the ideal 24bit solution that saved the intended quality and DXT5 is 32 bit with an alpha channel which I was warned about not to use unless my texture had an alpha included because this format is double the memory of DXT3!
  • CrazyButcher
    Offline / Send Message
    CrazyButcher polycounter lvl 18
    dxt3 and dxt5 are same size. just look at the wiki link has the details. [if an admin can do, we might want to break the trheads into two, splitting away the compression stuff]
  • System
    Offline / Send Message
    System admin
    That's strange, it was definately double the size after x conversion from tga inside the game editor as I extracted the textures to compare the dxt3 and dxt5 formats. This must be specific to the editor I was using, so if you use no alpha information in your dxt5 in some other editor the file size remains the same? Information noted:thumbup:
  • demoncage
    Offline / Send Message
    demoncage polycounter lvl 18
    I remember talking to my bro about this cause I didn't see the point. He brought up using it as a hack to get the superficial sss effect, which made sense to me. So I can see on stuff like skin or other materials with noticeable translucency that it could be effective but not a huge difference.

    But on everything else it seems like the additive effect is mimicing what you see on real world objects, and with the degree of specular control you can get w/ different maps, I'm not sure I see the returns in spending alot of time on color values. But whatever works for the indivdual artist.

    I know one horrendous experience I had, the client wanted me to not only tint the spec in a certain way on every material, but also to hand paint complex spec maps with stagnant hilights and duplicate that map for falloff control. Tell me how that makes ANY sense.

    using areas of grey for both strength and falloff looked great in the engine to me...
  • jeffdr
    Offline / Send Message
    jeffdr polycounter lvl 11
    DXT1 and DXT5 compress color in the exact same way; you won't see a quality difference for color data by switching between the two. DXT5 is bigger because it also holds alpha channel information, while DXT1 essentially does not (tho there are ways to use DXT1 and get a single bit of alpha, 0 or 1, by sacrificing a little color precision, but this is rarely done afaik).
  • EarthQuake
    Bal wrote: »
    Very interesting thread guys, and yeah I often heard the "use blue cause of the sky" explanation, which always sounded kind of silly.


    Just to touch on a specific point here a bit. Not responding specifically to your comment, it just reminded me.

    Generally if you can do it, its best to create your assets with a neutral lighting environment, something without a whole lot of color, but still representational of your game engine. Unless of course you know with certainty that the sky/lighting will always be the same color. The problem you get into is that you're color correcting for whatever environment you are previewing textures, and then in another environment stuff falls apart. Its important to let the lighting add the sky ambient/sun color to your assets(if you lighting system has those sort of things, of course).

    We had this problem quite a bit on darkest of days, where our artists would create content for the default sky that is loaded, make it look good, but then when its loaded up in a level with a different sky, its all whacked out(too much contrast, too little, crazy spec colors, etc).

    Other things like previewing your textures in anything other than your game engine, or a shader synced up to your engine's lighting in your 3d app of choice should be avoided at all costs as well. Especially if you have a really simple point light shader that you're texturing with in max, and then throwing that content into a nice engine with fancy image based lighting. You really need to texture to the specs of the engine you're using.
  • jeffdr
    Offline / Send Message
    jeffdr polycounter lvl 11
    Ok, so after poking around and doing some reading and thinking on the problem, I'm starting to think my initial guess was right - that this is an issue of gamma correction. This means two things:

    #1 - You artist types arent just smoking crack or anything :P In fact actually whoever originally picked up on the whole "use a cool spec map on warm materials" approach was pretty observant.

    #2 - This is something a programmer can fix for you. The root of the problem has to do with the assumptions made about texture inputs for lighting and compositing done by the author of your game, or your art tool.

    The problem is in fact quite widespread - many game engines and many art tools (like Photoshop) and even offline renderers (Max, Maya, others) have this problem, at least in their default configurations. I know the renderer I wrote (the one that drives Toolbag) has this problem too, because we stumbled on it too late in our project to fix it. So what's the deal, you ask?

    A bit of science first. A dielectric material (a non-conductive material), will not reflect light differently for different wavelengths. Examples of dielectric materials are wood, rock, dirt, skin, water, glass, and polymers such as plastic. Pretty much everything except for metal. This is not to say they don't diffuse light differently (clearly they do, they vary in color obviously!), but the actual reflection ("specularity") of light is not tinted. That is to say nearly everything you see IRL on a day to day basis has a white or at least grayscale spec map. Yes, really.

    So what's with all the colored spec maps out there? It was this question, paired with EQ's detailed examples, that got me thinking on this in the first place.

    The answer I think is gamma correction of input textures as well as the final displayed image. Adding two uncorrected images together doesn't work properly, and most importantly CAN ALTER THE HUE OF THE RESULT. What you'd expect if you added say, white to brown, is a less saturated brown. What you'd get instead without gamma correction would be sort of ass yellow/orange - *exactly* the kind of problem people are trying to get around by tinting their spec maps.

    Gamma correction is a mostly mathematical topic that I won't go into in detail here, but all of you who want to understand should at least skim through this article on the subject:

    http://http.developer.nvidia.com/GPUGems3/gpugems3_ch24.html

    Note in particular figure 24-5, that's about the most obvious example of improper gamma correction I've ever seen. If you've ever seen one of your models start to look like (b), when you really wanted (a), this is probably why. To be clear: the problem is purely technical, and very widespread. Tinting of spec maps is merely a workaround - properly addressing gamma on the rendering side will resolve this artifact and many others to boot.

    Other symptoms include artifacts with lighting, with generating mipmaps, with compositing of all kinds (such as adding spec on top of diffuse), and bad alpha blending. Basically everything goes just a little bit wrong. It can be hard to spot in most cases.

    As an example, for lighting, the process should be:

    (1) gamma correct the input textures
    (2) perform lighting calculation
    (3) un-gamma correct the result for proper display

    Often when people get it wrong, they leave off steps 1 and 3. Steps 1 and 3 mostly cancel each other out, but not completely, so it can be hard to notice in many instances when it's off.

    I hope this makes at least some sense, and that everyone goes forth today with a better understanding of their tools and their work.

    Cheers,

    Jeff
  • EarthQuake
    Just posting some of the images from that article

    24fig04.jpg
    24fig05.jpg
  • 00Zero
    thanks for taking the time to write this up guys. learned a lot.
  • SHEPEIRO
    Offline / Send Message
    SHEPEIRO polycounter lvl 17
    mmm most interesting, my weekend addled brains finding it a little difficult to assimilate though, isnt a screen gamma correctionx2 much more expensive than tweaking the colours of the specular, how much more accurate would it be to do it like this?
  • EarthQuake
    Well apparently OGL and DX9 both have functions to do this for free, or so i've been told.
  • glib
    So am I reading this right, that colored specs are just trying to cover for issues caused by bad rendering?
  • EarthQuake
    Not in all cases, but potentially some, yes.

    Some materials really do need a different spec color to represent properly.
  • EarthQuake
    Just to touch on some of this compression stuff a bit more, so we've come to the conclusion that using a greyscale spec, either in the alpha or as a separate image doesnt actually save you any memory.

    So why would anyone want to do it?

    Well there are a couple reasons:

    1. In terms of performance, grabing a texture from the alpha channel is going to be quicker, and result in a simplier shader(code wise). This may have barely any effect on 1 mesh, but if you use the same shader for virtually every mesh in the game, it could have a decent hit on performance. I also think that you can squeeze in more inputs by throwing stuff into alphas, because each card has a set # of texture calls it can do, or something of that nature. Also, if you're working with older hardware, it may be your only option, because of the texture call thing.

    2. If your game uses uncompressed textures, then you actually will see some memory savings for sticking things in the alpha, or saving as 8 bit images.

    3. Just good old fashion miss-information. =)
  • jeffdr
    Offline / Send Message
    jeffdr polycounter lvl 11
    EQ has asked me to lay some programmer shit on you guys regarding texture storage in real time renderers, so here goes.

    Some common textures formats:
    1. DXT1: 4 bits per pixel, typically RGB only
    2. DXT5: 8 bits per pixel (4 for color and 4 for alpha)
    3. L8: 8 bits per pixel, uncompressed grayscale
    4. R8G8B8: in theory 24 bits per pixel, but often stored to use 32 bits for memory alignment reasons
    5. R8G8B8A8: full 32 bits per pixel, with alpha channel
    There are of course many more formats than these 5, but these are the "all stars" so to speak; a vast majority of texture data is stored in one of these layouts. So these are the typical puzzle pieces we have to work with when deciding what textures to put where, how to arrange data, etc.

    From these numbers you can see that using for example a dxt5 texture is going to take up twice the space as a dxt1, and all you get in the trade is a low precision alpha channel. In terms of quality it may be slightly better to use two dxt1 images (I'll leave that to you to judge), which would be equivalent in terms of space. Plus then you can use color.

    You'll also notice that an uncompressed grayscale image takes the same amount of space as a compressed RGBA image in dxt5 format. Take from that what you will.

    Also worth noting is that the 24bit RGB format is often stored as 32bit with an unused alpha channel (for purely technical reasons), and so if you are using uncompressed textures often that alpha channel is totally free - the memory is already used so you can go ahead and drop stuff in there at no cost (except maybe disk space / load time).

    Now with all that said, it is still possibly in your interest to use as few textures as possible. If you have two shaders that are equivalent, and one uses say 4 textures when the other version uses 3, your programmers are probably going to tell you that the version using 3 is preferable. There are two reasons for this:
    1. Driver overhead. It takes some time and work on the CPU to switch textures, or to ready them for use. This overhead is small, and the guys at ATI/NVIDIA et al take great pains to minimize it, but it is still there. So you will lose a bit of CPU time using more textures. This varies only with the number of textures, not their size or format.
    2. Rendering cost. The more textures your graphics card has to sample, the slower it will render. The cost is again not drastic, but it is measurable in many cases. If you have a shader that uses *lots* of textures (in DoD we have one that uses 14!), this cost can really start to build. So this is a drain on the GPU. This cost does vary with the size of the textures.
    So this is just sort of a technical overview; you will have to decide based on your tools/pipeline/goals (and programmer skill :P) what works best for you.

    Jeff
  • Bal
    Offline / Send Message
    Bal polycounter lvl 17
    Good info guys, thanks for sharing this.
  • glib
    EarthQuake wrote: »
    Just to touch on some of this compression stuff a bit more, so we've come to the conclusion that using a greyscale spec, either in the alpha or as a separate image doesnt actually save you any memory.

    So why would anyone want to do it?

    Well there are a couple reasons:

    1. In terms of performance, grabing a texture from the alpha channel is going to be quicker, and result in a simplier shader(code wise). This may have barely any effect on 1 mesh, but if you use the same shader for virtually every mesh in the game, it could have a decent hit on performance. I also think that you can squeeze in more inputs by throwing stuff into alphas, because each card has a set # of texture calls it can do, or something of that nature. Also, if you're working with older hardware, it may be your only option, because of the texture call thing.

    2. If your game uses uncompressed textures, then you actually will see some memory savings for sticking things in the alpha, or saving as 8 bit images.

    3. Just good old fashion miss-information. =)

    4. There's a reason to use a 1-bit spec in your scene, or perhaps a mask for reflection or something else. In this case, dropping it into a DXT1 compressed texture is (AFAIK) free.
  • Whargoul
    Offline / Send Message
    Whargoul polycounter lvl 18
    I was reading through the thread, writing up my response in my head, thinking all about gamma-correct lighting and dielectric materials, and lo and behold jeffdr posts it for me :)

    But yeah, to reiterate: the bad tinting you get with a neutral coloured spec is due to doing your lighting in a non-gamma corrected way. Get your render engine modified do it right (be prepared to redo all the lighting though...). You can still get colour shifting though - due to clipping. Even if you are rendering HDR, when you bring your colours back to LDR space - some channels may clip before other ones do - leading to a colour shift. This is a nasty one, and tough to get rid off with a nice visual quality.

    Specular highlights should always be the colour of the input light source (since it is really a reflection of that light source) unless it is a specific type of metallic object, in which case the surface tints it accordingly (ie gold really does have gold-tinted highlights).

    Another reason to use greyscale spec maps is that they can be packed together in a single texture with other single channel maps. Yes, adding them to the alpha is a sucky way to do it since you are effectively doubling the memory. If you have 3 maps, each with an alpha channel, you can rip those alphas out, pack them together, and have 4 maps total which uses 2/3's the memory of the 3 with alpha. But one reason to leave it in the alpha is due to sampler input limits on some hardware, ie it can only access 8 maps at a time so then leaving it in the alpha has a benefit. And as jeffdr said, less textures = better performance.

    In our current engine, it renders gamma correct so we have no tinting problems, except for the very bright clipping problem at the moment. We pack our greyscale specular map, specular power (also called eccentricity, gloss, roughness, etc), and ambient occlusion into a single map - so base shaders use difffuse, normal and the packed map - we also have an RGB value per material for globally tinting the spec highlight for special surfaces. I am not currently tinting the spec very often, and certainly not using blue on the skin, as it had the effect of making the skin too cold and harsh, neutral was making it softer. Of course, the fast SSS fake helps warm it up as well.

    14 textures on one object? heh, I had one using 22 textures :) Although some of those are "system" textures like the shadow map, env map, noise maps that all shaders share, but yeah it was crazy. Ran on PC & 360 no problem, but the PS3 choked because of sampler limits (I think only 16?). Anyways, it was a pre-pro shader and ended up being cleaned up - to 16 exactly.
  • EarthQuake
    Just curious, why are you loading AO as a separate image and not baked into the diffuse/spec?
  • MattBush
    Realistically the AO should not affect direct light sources, only ambient light, and diffused ambient reflections.

    See this quality ms paint diagram:

    aodiagram.png
    Notice how direct light and specular reflections are only affected by the amount the sample is in shadow from the direct light source, and not by any global/ambient function.

    What I do, is use the AO to modulate the ambient light (from an ambient cube/light probe), then for reflections, use the gloss map to select the mip level of the reflection cube and modulate the reflections with the AO more/less according to how glossy the sample is. So more diffused reflections get affected by the AO more than glossy ones.
    And that is the extent to which the AO should be used in my opinion.

    You may of course prefer the look of AO in the diffuse and spec, but it's something to consider.

    Also, hi!
  • Brice Vandemoortele
    Offline / Send Message
    Brice Vandemoortele polycounter lvl 19
    Hi mattbush, welcome and thx for those explanations :)
    It reminds me of talks around the SSAO buzz: most studios use it as a post process multiplied on top of the whole back buffer. Some studios on the other hand choose to output the ssao in a texture in screenspace sampled by the shaders only to be used to modulate the ambient lighting contribution (i think i read that about crysis but i might be wrong).
    I had the same experience not so long ago when using ao maps on a separate uv set. Even when gamma corrected, the specular looks pretty wrong when it's multiplied by ao.
    (glad to see other people use glossy cubemaps :P, interesing use of ao AND specPow)
  • MoP
    Offline / Send Message
    MoP polycounter lvl 18
    EQ: Yeah, if you multiply AO into diffuse then it will always shadow even if the surface is lit by a direct light (which doesn't happen in reality) ... it's called "ambient occlusion" for a reason!

    If you have it as a separate map then you can actually use it in a more correct way - only multiply if not affected by direct light (or maybe a range of values to multiply so a strong light source will completely blast away the AO while a weaker light source will show the AO through a bit).

    Obviously this is more expensive in terms of texture memory and shader complexity though, which is why most people don't do it currently. I imagine it will become more common as hardware gets more powerful.

    MattBush: Thanks for posting, interesting stuff! I never considered using the gloss to select the cubemap mip level, that's pretty cool.
  • [Deleted User]
    Offline / Send Message
    [Deleted User] polycounter lvl 18
    This information is really neat. This might be a really silly question, but is proper gamma correction something that an artist can accomplish in a shader, or is it the kind of thing that would be hard-coded into the rendering engine?

    Also, somewhat relatedly (though this is probably a little subjective, and probably even sillier of a question), if it can be done with a shader, reducing/eliminating the need for colored specs on most materials, would it be a safe bet to go that route for portfolio pieces or would it be better to do it the other way so it's clear one knows how to do it? Or would it even matter much either way? I guess it's minor, but I'm just trying to get an idea so I can do whatever's likely to be more well-received. Like, if it was gamma-corrected and it rendered well using grayscale spec for human skin, is that something that, in an applicant with no experience, would look cool for being tech-savvy, or would it just be taken as "not knowing to make them blue" or something?
  • Whargoul
    Offline / Send Message
    Whargoul polycounter lvl 18
    Yeah, what those guys said. I use the AO in many ways:
    *full strength vs ambient light (duh)
    *maybe %20 or so vs direct lights (still adds a bit of shading, without looking like a big multiply)
    *contasted & brightened vs reflections (kill refs in dark spots)
    *same again with different params vs ambient specular (helps weaken bright spec from sky, etc)

    Shader complexity is pretty minor, a couple lerps, multiplies and adds. Packing the values together and doing fancy math is they key :)

    Yes, do NOT use SSAO as a full screen multiply - yuck. That's as bad as doing a full screen multiply of your shadow map!

    Our current AO is a combination of a few things, like geometry baked AO, screen space AO, textural AO, etc.
  • EarthQuake
    My biggest concern is that for certain materials, flesh for instance, you dont want to have grayscale AO multiplied on, you want a more saturated pink sort of color. But maybe this too would be improved with proper gamma correction?

    And as far as personal preference, i think it helps to make what are generally flat surfaces with much more detail normal mapped onto them, pop out more and look better to have that ao baked into the diff/spec. That said i havent really seen any examples of what you guys are talking about here either. At first glance i would see having it in a separate map with a custom shader benifiting lightmaps on large environment pieces moreso than unique AO baked from the highres for characters and props, but what do i know.

    Anyone have some realworld examples?
  • Whargoul
    Offline / Send Message
    Whargoul polycounter lvl 18
    Well it doesn't look bad on skin because it's not multiplied on at any time, it just removes light.
  • MoP
    Offline / Send Message
    MoP polycounter lvl 18
    Yeah I think we will always need to keep the AO baked into the diffuse until self-shadowing normal-maps become the standard.

    Since currently all you really know is whether the direct light is hitting the pixel or not, but this doesn't take into account any "overlap" or "protruding" detail you want in the normal-map, unless you use a self-shadowing shader.

    So until they become standard I think the AO should at least always be at least a bit evident in the diffuse & specular, otherwise you will find your textures losing contrast.

    Anyway, it's obviously the case that this stuff pretty much only applies if you're looking for realism - if you want a more stylised look then you may even want to crank up the AO baked into your diffuse/spec for more visual "pop".
  • EarthQuake
    Whargoul wrote: »
    Well it doesn't look bad on skin because it's not multiplied on at any time, it just removes light.

    Oh alright, so i imagine that if you had say, a game that was almost entirely outdoor lighting with very little direct lighting, it wouldn't really be worth it to do?
  • MoP
    Offline / Send Message
    MoP polycounter lvl 18
    No, it would, because in outdoors you have a very strong direction light source (the sun), and the rest is ambient. So in this case the ambient occlusion map would be fully visible on the "un-lit" side (ie. a character's back, if they are facing into the sun), and the ambient occlusion map would not be used on their front since it was lit directly.
  • EarthQuake
    Oh yeah, right. derr
  • Firebert
    Offline / Send Message
    Firebert polycounter lvl 15
    screw this amazing informational technical data...i am most impressed by EQ's well manicured hands.
1
Sign In or Register to comment.